Research On Disease Diagnosis Model Based On Machine Learning

Health is the basis for the all-round development of mankind.In recent years,the incidence of various diseases has shown an increasing trend,especially in areas where medical treatment is underdeveloped,which undoubtedly makes the work of doctors more difficult.The research of disease diagnosis based on machine learning is of great significance to improve the accuracy and real-time of disease diagnosis,and to assist doctors in completing disease diagnosis.The machine learning algorithm learns the data generated by medical institutions to obtain a disease diagnosis model,which can predict unknown data samples.However,it is difficult for the traditional single-classifier diagnosis model to obtain excellent generalization ability.Therefore,in the application of machine learning algorithms,it is also necessary to consider the integration and optimization of various technologies.Based on the study of the disease diagnosis model based on traditional machine learning,the paper focuses on the multi-classifier integrated diagnosis model and the multi-classifier selective integrated diagnosis model.The main research includes:Firstly,a Bagging-Adaboost-SVM multi-classifier ensemble diagnosis model is proposed.First,the two classification methods of fusion of Bagging,Adaboost and SVM three algorithms,using the idea of Bagging algorithm,each time only a small part of the data is extracted from the disease sample as the training set;secondly,each sub-training set adopts the denoising algorithm to find the noise Data and modify its class label according to the statistical data of its category;then,Adaboost-SVM is trained on each sub-training set and the number of weak classifiers is filtered(determined according to the accuracy of the weak classifier);Finally,integrate the weak classifiers to obtain the final classification model of each sub-data set as the base classifier of Bagging.Experimental results show that the accuracy of the model on the diabetes data set reaches 92.36%,which is 3.16% higher than the deep neural network(DNN)algorithm used in diabetes diagnosis,which verifies that the model is more suitable for diabetes diagnosis.Secondly,the BDA-GA multi-classifier selective ensemble diagnosis model is proposed.First of all,use the re-sampling technique with replacement to select training samples from the original data set,and generate several base classifiers through the classification algorithm and the training set;second,use the weighting method to balance the diversity between the classifiers and the accuracy of the classifiers,Namely BDA;then,train based on the validation set and the base classifier and calculate the BDA value,use the selection algorithm(GA)to search for the optimal base classifier subset according to the BDA evaluation criteria,and finally,the obtained high-quality subset Use the voting method to integrate the diagnosis results to get the final result.Experimental results show that the accuracy of the model on the hypertension data set reaches 93.83%,which is 4.27% higher than the artificial neural network(ANN)algorithm suitable for hypertension diagnosis.